US8392436B2 - Semantic search via role labeling - Google Patents

Semantic search via role labeling Download PDF

Info

Publication number
US8392436B2
US8392436B2 US12364041 US36404109A US8392436B2 US 8392436 B2 US8392436 B2 US 8392436B2 US 12364041 US12364041 US 12364041 US 36404109 A US36404109 A US 36404109A US 8392436 B2 US8392436 B2 US 8392436B2
Authority
US
Grant status
Grant
Patent type
Prior art keywords
query
semantic
network
neural
computer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active, expires
Application number
US12364041
Other versions
US20090204605A1 (en )
Inventor
Bing Bai
Jason Weston
Ronan Collobert
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
NEC Corp
Original Assignee
NEC Laboratories America Inc
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Grant date

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/30Information retrieval; Database structures therefor ; File system structures therefor
    • G06F17/3061Information retrieval; Database structures therefor ; File system structures therefor of unstructured textual data
    • G06F17/30634Querying
    • G06F17/30637Query formulation
    • G06F17/30654Natural language query formulation or dialogue systems

Abstract

A method and system for searching for information contained in a database of documents each includes an offline part and an online part. The offline part includes predicting, in a first computer process, semantic data for sentences of the documents contained in the database and storing this data in a database. The online part includes querying the database for information with a semantically-sensitive query, predicting, in a real time computer process, semantic data for the query, and determining, in a second computer process, a matching score against all the documents in the database, which incorporates the semantic data for the sentences and the query.

Description

RELATED APPLICATION

This application claims the benefit of U.S. Provisional Application No. 61/026,844, filed Feb. 7, 2008, the entire disclosure of which is incorporated herein by reference.

FIELD OF THE INVENTION

This invention relates to query answering systems. More particularly, this invention relates to a query answering system that uses semantic searching methods.

BACKGROUND OF THE INVENTION

The marriage of Natural Language Processing (NLP) with Information Retrieval (IR) has long been desired by researchers. NLP techniques are intensely used in query answering (QA) systems, indeed QA has been a playground for Artificial Intelligence since the 1960s. NLP techniques such as ontologies, syntactic parsing and information extraction techniques can be commonly found in a good QA system.

Although QA has been successful in domain specific areas and in small document collection such as TREC, large scale open-domain QA is still very difficult because the current NLP techniques are too expensive for massive databases like the internet Therefore, some commercial systems resort to simplified NLP, where, for example an attempt is made to map user queries (questions) to previously hand-picked question/answer sets. Thus, the performance of such systems is limited to their QA database.

Recently, a fast convolutional neural network approach for semantic extraction called SENNA has been described, which achieves state-of-art performance for Propbank labeling, while running hundreds of times faster than competing methods.

SUMMARY

A method is disclosed herein for searching for information contained in a database of documents. The method comprises predicting, in a first computer process, semantic data for sentences of the documents contained in the database, querying the database for information with a semantically-sensitive query, predicting, in a real time computer process, semantic data for the query, and determining, in a second computer process, a matching score against all the documents in the database, which incorporates the semantic data for the sentences and the query.

Also disclosed herein is a system for searching for information contained in a database of documents. The system comprises a central processing unit and a memory communicating with the central processing unit. The memory comprises instructions executable by the processor for predicting, in a first computer process, semantic data for sentences of the documents contained in the database, querying the database for information with a semantically-sensitive query, predicting, in a real time computer process, semantic data for the query, and determining, in a second computer process, a matching score against all the documents in the database, which incorporates the semantic data for the sentences and the query.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a flowchart illustrating an exemplary embodiment of the semantic searching method of the present disclosure.

FIG. 2 is a flowchart illustrating an exemplary embodiment of a neural network tagger of the present disclosure.

DETAILED DESCRIPTION

Disclosed herein is a semantic searching method for use in a query answering (QA) system. The semantic searching method efficiently indexes and retrieves sentences based on a semantic role matching technique that uses a neural network architecture similar to SENNA. First, semantic roles (metadata) are computed offline on a large database of documents (e.g., general web data used by search engines or special collections, such as Wikipedia) and the metadata is stored along with word information necessary for indexing. At query time, semantic roles are predicted for the query online and a matching score against all the documents in the database, is computed that incorporates this semantic information.

The offline processing and online labeling are performed using fast prediction methods. Therefore, given the indices computed offline, the retrieval time of the present system is not much longer than other simple IR models such as a vector space model, while the indexing itself is affordable for daily crawling. More importantly, the neural network framework described herein is general enough to be adapted to other NLP features as well, such as named entity recognition and coreference resolution.

FIG. 1 is a flowchart illustrating an exemplary embodiment of the semantic searching method of the present disclosure. The method generally comprises an offline process and an online process.

In the offline process, documents are collected in step 10 and processed in step 13 with a sentence splitter. The sentence splitter separates each document into discrete sentences. The sentences at the output 14 of the sentence splitter, are processed by a tagger in step 11. In the present disclosure, the tagger comprises a neural network. The neural network tagger computes part-of speech (POS) tags and semantic role tag predictions for each sentence. The sentences and their POS tags and role labeling information at the output 15 of the neural network tagger are indexed at step 101 and stored in a database in step 19 as a forward index and an inverted index.

In the online process, a user query 17 entered into browser 20 is sent by web server 30 to the neural network tagger for processing on the fly. The neural network tagger computes part-of speech (POS) tags and semantic role tag predictions for the query. In step 12, the web server calculates the similarity between the query and the sentences stored in the database containing the forward and inverted indices calculated during the offline process using the query's POS tags and role labeling information at the output 16 of the neural network tagger and each sentence's syntactic and/or role labeling information stored in the database as forward and inverted indices, and then ranks the similarity calculations. The result of the similarity ranking is displayed by the browser 20 for viewing by the user in step 18. The result comprises top ranked sentences which are most likely to match the semantic meaning of the query 17.

FIG. 2 is a flowchart illustrating an exemplary embodiment of the neural network tagger. The neural network tagger comprises but is not limited to a five layer convolutional neural network. The neural network tagger provides a semantic role prediction for a single given word of interest. This procedure has to be repeated for each word in the sentence/query. For ease of description and not limitation, the neural network tagger will now be described using a word of interest w that corresponds to a verb v and the sentence/query, “The company operates stores mostly . . . .” For each word input to the network, two salient features are: (1) its distance from the word w being labeled; and (2) its distance from the verb v of interest. These two features, and the words themselves, are the only input to the neural network tagger. Because, the neural network tagger cannot handle text directly, in step 21, the first layer of the neural network tagger converts each word in the input sentence via a lookup table (dictionary) into a low dimensional (e.g., 20 dimensional) vector that comprises the features that represent that word. These features are learned during backpropagation. If the input sentence contains n words, the lookup table provides a 20×n feature representation. Two more types of features are also concatenated onto those features, the distance to the word w to label and the distance to the verb v. Both of these features are also encoded via lookup tables (e.g., each 5 dimensional). The neural network tagger will learn its own measure of distance useful for obtaining good classification accuracy.

In the second layer of the neural network tagger, represented by steps 22 and 23, the vectors for each word in the sentence/query are transformed by convolution to a matrix (step 22) which includes, for example, k+1 columns (vectors) and 200 rows (step 23). Using this example, the convolutional second layer is capable of outputting 200 features for every window of, for example, 3 adjoining words in the sentence/query. The convolutional second layer enables the neural network to deal with variable length sentences/queries. So for a sentence/query of length n the second layer of the neural network tagger outputs (n−2)×200 features.

In the third layer of the neural network tagger, the columns or vectors (each of which comprises 200 rows) of the resulting matrix from the second layer are converted into a single vector in step 24 by examining each row of the columns to find the largest (maximum) value, and constructing the single vector from the maximum value in each row of the columns. The max function is applied over that feature set with the intuition that the most pertinent locations in the sentence for various learned features are identified at this layer of the neural network tagger. Independent of sentence/query length, this exemplary embodiment of the neural network tagger outputs 200 features. The single vector is processed by the fourth and fifth layers of the neural network tagger in step 25 to get the role label of the word w. Specifically in this exemplary embodiment, 100 hidden units are applied to the single vector in the fourth layer of the neural network and the fifth layer of the neural network predicts possible outputs (classes). The fifth layer may be a linear layer that predicts 24 possible classes. The neural network tagger is trained by backpropagation to minimize training error.

In an exemplary embodiment, the similarity between the query and the sentences stored in the forward and inverted indices may be calculated by the web server during the offline process using the following method. First, the similarity between a query Q and a sentence S may be represented as Equations (1) and (2).

SENmatch ( Q , S ) = a PAS ( Q ) w ( a ) max b PAS ( S ) PASmatch ( a , b ) , w ( a ) = idf ( v a ) k PAS ( Q ) idf ( v k ) ( 1 ) PASmatch ( a , b ) = V semtfidf T ( a ) · V semtfidf ( b ) V semtfidf T ( a ) V semtfidf ( a ) V semtfidf T ( b ) V semtfidf ( b ) · I ( a , b ) , I ( a , b ) = { 1 , v a and v b are synonyms 0 , otherwise ( 2 )

A predicate-argument-structure (PAS) of a sentence, is the part of sentence which has a meaningful role for a certain verb. For example, the sentence, “the cat sat on the mat and ate fish” has two PASs, i.e., 1) “the cat sat on the mat”; 2) “the cat ate fish”. For each PAS a in query Q, similarity is calculated with all PASs in sentence S, and consider the maximum as the similarity between a and S. The similarity SENmatch(Q,S) is a weighted sum of similarities between each query PAS and S. The weight is the relative importance of the verb in the query PAS, indicated by its inverse document frequency (IDF).

The similarity between PAS a and b (PASmatch(a,b)) is a variant of classical cosine measure, with two modifications. In the first modification, the vector Vsemtfidf(x) for a PAS x is a vector of size |W∥R|, where |W| is the size of vocabulary, and |R| is the number of possible semantic roles. In other words, the vector has an entry for each combination of a word and a semantic role. If there is a word w (w is a unique integer id of a word, starting with 0) with a role r (r is a unique integer id for a semantic role, starting with 0) in the PAS x, then the entry with index (w|R|+r) is set to the IDF of word w; otherwise the entry is set to 0. In the second modification, the similarity between two PASs is considered only if their verbs are synonyms, as indicated by the term I(a,b) in Eq. (2)

One skilled in the art will recognize that both the web server and the computer system that performs the offline process, as described herein, may each comprise any suitable computer system. The computer system may include, without limitation, a mainframe computer system, a workstation, a personal computer system, a personal digital assistant (PDA), or other device or apparatus having at least one processor that executes instructions from a memory medium.

The computer system may include one or more memory mediums on which one or more computer programs or software components may be stored. The one or more software programs which are executable to perform the methods described herein, may be stored in the memory medium. The one or more memory mediums may include, without limitation, CD-ROMs, floppy disks, tape devices, random access memories such as but not limited to DRAM, SRAM, EDO RAM, and Rambus RAM, non-volatile memories such as, but not limited hard drives and optical storage devices, and combinations thereof. In addition, the memory medium may be entirely or partially located in one or more associated computers or computer systems which connect to the computer system over a network, such as the Internet.

The methods described herein may also be executed in hardware, a combination of software and hardware, or in other suitable executable implementations. The methods implemented in software may be executed by the processor of the computer system or the processor or processors of the one or more associated computers or computer systems connected to the computer system.

While exemplary drawings and specific embodiments of the present invention have been described and illustrated, it is to be understood that that the scope of the present invention is not to be limited to the particular embodiments discussed. Thus, the embodiments shall be regarded as illustrative rather than restrictive, and it should be understood that variations may be made in those embodiments by workers skilled in the arts without departing from the scope of the present invention as set forth in the claims that follow and their structural and functional equivalents.

Claims (3)

1. A method for searching for information contained in a database of documents, the method comprising the steps of:
in a computer process, predicting parts-of-speech and semantic tags for sentences of the documents in the database;
in a computer process, storing the sentences and the tags in an inverted index;
requesting information from the database with a query;
in a computer process, predicting parts-of-speech and semantic tags for the query;
in a computer process, determining a score indicative of the similarity between each of the sentences and the query using the tags corresponding to the sentences stored in the index and the tags corresponding to the query, wherein the similarity between each of the sentences and the query comprises:
SENmatch ( Q , S ) = a PAS w ( a ) max b PAS ( S ) PASmatch ( a , b ) , w ( a ) = idf ( v a ) k PAS ( Q ) idf ( v k ) and PASmatch ( a , b ) = V semtfidf T ( a ) · V semtfidf ( b ) V semtfidf T ( a ) V semtfigf ( a ) V semtfidf T ( b ) V semtfidf ( b ) · I ( a , b ) , I ( a , b ) = { 1 , V a and V b are synonyms 0 , otherwise ,
and wherein Q is the query, S is the sentence, SENmatch(Q, S) is the similarity between the query and the sentence, PAS is a predicate-argument structure of the sentence, idf is inverse document frequency of verb v, PASmatch(a, b) is the similarity between PAS a and PAS b, and Vsemtfidf is a vector of PAS;
in a computer process, ranking the sentences according to their corresponding scores, wherein the sentences with the highest rankings match the semantic meaning of the query.
2. The method of claim 1, wherein the steps of predicting the parts-of-speech and semantic tags for the sentences and storing the sentences and the tags in the inverted index, are performed offline.
3. The method of claim 1, wherein the steps of predicting the parts-of-speech and semantic tags for the query, determining the similarity scores, and ranking the sentences, are performed online.
US12364041 2008-02-07 2009-02-02 Semantic search via role labeling Active 2029-12-27 US8392436B2 (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US2684408 true 2008-02-07 2008-02-07
US12364041 US8392436B2 (en) 2008-02-07 2009-02-02 Semantic search via role labeling

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
US12364041 US8392436B2 (en) 2008-02-07 2009-02-02 Semantic search via role labeling

Publications (2)

Publication Number Publication Date
US20090204605A1 true US20090204605A1 (en) 2009-08-13
US8392436B2 true US8392436B2 (en) 2013-03-05

Family

ID=40939774

Family Applications (1)

Application Number Title Priority Date Filing Date
US12364041 Active 2029-12-27 US8392436B2 (en) 2008-02-07 2009-02-02 Semantic search via role labeling

Country Status (1)

Country Link
US (1) US8392436B2 (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130204611A1 (en) * 2011-10-20 2013-08-08 Masaaki Tsuchida Textual entailment recognition apparatus, textual entailment recognition method, and computer-readable recording medium
US9477654B2 (en) 2014-04-01 2016-10-25 Microsoft Corporation Convolutional latent semantic models and their applications
US9519859B2 (en) 2013-09-06 2016-12-13 Microsoft Technology Licensing, Llc Deep structured semantic model produced using click-through data
US9535960B2 (en) 2014-04-14 2017-01-03 Microsoft Corporation Context-sensitive search using a deep learning model
US9552549B1 (en) 2014-07-28 2017-01-24 Google Inc. Ranking approach to train deep neural nets for multilabel image annotation

Families Citing this family (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9092425B2 (en) * 2010-12-08 2015-07-28 At&T Intellectual Property I, L.P. System and method for feature-rich continuous space language models
WO2014040263A1 (en) * 2012-09-14 2014-03-20 Microsoft Corporation Semantic ranking using a forward index
US9239872B2 (en) * 2012-10-11 2016-01-19 Nuance Communications, Inc. Data store organizing data using semantic classification
US9081847B2 (en) 2012-10-11 2015-07-14 Nuance Communications, Inc. Data store organizing data using semantic classification
US9613125B2 (en) 2012-10-11 2017-04-04 Nuance Communications, Inc. Data store organizing data using semantic classification
US9069857B2 (en) * 2012-11-28 2015-06-30 Microsoft Technology Licensing, Llc Per-document index for semantic searching
US20150066496A1 (en) * 2013-09-02 2015-03-05 Microsoft Corporation Assignment of semantic labels to a sequence of words using neural network architectures
CN103514883B (en) * 2013-09-26 2015-12-02 华南理工大学 An adaptive method of switching the sound men and women
US9400925B2 (en) * 2013-11-15 2016-07-26 Facebook, Inc. Pose-aligned networks for deep attribute modeling
KR101787062B1 (en) * 2014-01-07 2017-10-19 한국전자통신연구원 Apparatus and Method for searching information based on Wikipedia's contents
US20150317302A1 (en) * 2014-04-30 2015-11-05 Microsoft Corporation Transferring information across language understanding model domains
WO2016024261A1 (en) * 2014-08-14 2016-02-18 Opisoftcare Ltd. Method and system for searching phrase concepts in documents
CN104933031A (en) * 2015-06-29 2015-09-23 华东师范大学 Unsupervised automatic Q&A method based on semantic web
CN105183714A (en) * 2015-08-27 2015-12-23 北京时代焦点国际教育咨询有限责任公司 Sentence similarity calculation method and apparatus
US20170060844A1 (en) * 2015-08-28 2017-03-02 Microsoft Technology Licensing, Llc Semantically-relevant discovery of solutions
US9836671B2 (en) 2015-08-28 2017-12-05 Microsoft Technology Licensing, Llc Discovery of semantic similarities between images and text
US9910914B1 (en) * 2016-05-05 2018-03-06 Thomas H. Cowley Information retrieval based on semantics
WO2018018626A1 (en) * 2016-07-29 2018-02-01 Microsoft Technology Licensing, Llc. Conversation oriented machine-user interaction
US20180052929A1 (en) * 2016-08-16 2018-02-22 Ebay Inc. Search of publication corpus with multiple algorithms

Citations (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6189002B1 (en) * 1998-12-14 2001-02-13 Dolphin Search Process and system for retrieval of documents using context-relevant semantic profiles
US6240409B1 (en) * 1998-07-31 2001-05-29 The Regents Of The University Of California Method and apparatus for detecting and summarizing document similarity within large document sets
US6246977B1 (en) * 1997-03-07 2001-06-12 Microsoft Corporation Information retrieval utilizing semantic representation of text and based on constrained expansion of query words
US6269368B1 (en) * 1997-10-17 2001-07-31 Textwise Llc Information retrieval using dynamic evidence combination
US6480843B2 (en) * 1998-11-03 2002-11-12 Nec Usa, Inc. Supporting web-query expansion efficiently using multi-granularity indexing and query processing
US20040215663A1 (en) * 2001-11-30 2004-10-28 Microsoft Corporation Media agent
US20050080776A1 (en) * 2003-08-21 2005-04-14 Matthew Colledge Internet searching using semantic disambiguation and expansion
US6999959B1 (en) * 1997-10-10 2006-02-14 Nec Laboratories America, Inc. Meta search engine
US20070112764A1 (en) * 2005-03-24 2007-05-17 Microsoft Corporation Web document keyword and phrase extraction
US20070156669A1 (en) * 2005-11-16 2007-07-05 Marchisio Giovanni B Extending keyword searching to syntactically and semantically annotated data
US20080221878A1 (en) * 2007-03-08 2008-09-11 Nec Laboratories America, Inc. Fast semantic extraction using a neural network architecture
US20090112835A1 (en) * 2007-10-24 2009-04-30 Marvin Elder Natural language database querying
US7890539B2 (en) * 2007-10-10 2011-02-15 Raytheon Bbn Technologies Corp. Semantic matching using predicate-argument structure

Patent Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6246977B1 (en) * 1997-03-07 2001-06-12 Microsoft Corporation Information retrieval utilizing semantic representation of text and based on constrained expansion of query words
US6999959B1 (en) * 1997-10-10 2006-02-14 Nec Laboratories America, Inc. Meta search engine
US6269368B1 (en) * 1997-10-17 2001-07-31 Textwise Llc Information retrieval using dynamic evidence combination
US6240409B1 (en) * 1998-07-31 2001-05-29 The Regents Of The University Of California Method and apparatus for detecting and summarizing document similarity within large document sets
US6480843B2 (en) * 1998-11-03 2002-11-12 Nec Usa, Inc. Supporting web-query expansion efficiently using multi-granularity indexing and query processing
US6189002B1 (en) * 1998-12-14 2001-02-13 Dolphin Search Process and system for retrieval of documents using context-relevant semantic profiles
US20040215663A1 (en) * 2001-11-30 2004-10-28 Microsoft Corporation Media agent
US7283992B2 (en) * 2001-11-30 2007-10-16 Microsoft Corporation Media agent to suggest contextually related media content
US20050080776A1 (en) * 2003-08-21 2005-04-14 Matthew Colledge Internet searching using semantic disambiguation and expansion
US20070112764A1 (en) * 2005-03-24 2007-05-17 Microsoft Corporation Web document keyword and phrase extraction
US20070156669A1 (en) * 2005-11-16 2007-07-05 Marchisio Giovanni B Extending keyword searching to syntactically and semantically annotated data
US20080221878A1 (en) * 2007-03-08 2008-09-11 Nec Laboratories America, Inc. Fast semantic extraction using a neural network architecture
US7890539B2 (en) * 2007-10-10 2011-02-15 Raytheon Bbn Technologies Corp. Semantic matching using predicate-argument structure
US20090112835A1 (en) * 2007-10-24 2009-04-30 Marvin Elder Natural language database querying

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130204611A1 (en) * 2011-10-20 2013-08-08 Masaaki Tsuchida Textual entailment recognition apparatus, textual entailment recognition method, and computer-readable recording medium
US8762132B2 (en) * 2011-10-20 2014-06-24 Nec Corporation Textual entailment recognition apparatus, textual entailment recognition method, and computer-readable recording medium
US9519859B2 (en) 2013-09-06 2016-12-13 Microsoft Technology Licensing, Llc Deep structured semantic model produced using click-through data
US9477654B2 (en) 2014-04-01 2016-10-25 Microsoft Corporation Convolutional latent semantic models and their applications
US9535960B2 (en) 2014-04-14 2017-01-03 Microsoft Corporation Context-sensitive search using a deep learning model
US9552549B1 (en) 2014-07-28 2017-01-24 Google Inc. Ranking approach to train deep neural nets for multilabel image annotation

Also Published As

Publication number Publication date Type
US20090204605A1 (en) 2009-08-13 application

Similar Documents

Publication Publication Date Title
Wang et al. Multi-document summarization via sentence-level semantic analysis and symmetric matrix factorization
Büttcher et al. Information retrieval: Implementing and evaluating search engines
Harter Psychological relevance and information science
US6678677B2 (en) Apparatus and method for information retrieval using self-appending semantic lattice
Harabagiu et al. The role of lexico-semantic feedback in open-domain textual question-answering
US7305389B2 (en) Content propagation for enhanced document retrieval
US20080059187A1 (en) Retrieval of Documents Using Language Models
US7496548B1 (en) Neural network for electronic search applications
US20080195601A1 (en) Method For Information Retrieval
US20070136246A1 (en) Answer determination for natural language questioning
US20050234880A1 (en) Enhanced document retrieval
US20090292685A1 (en) Video search re-ranking via multi-graph propagation
US20140006012A1 (en) Learning-Based Processing of Natural Language Questions
US20120077178A1 (en) System and method for domain adaptation in question answering
US20070233656A1 (en) Disambiguation of Named Entities
US20040024756A1 (en) Search engine for non-textual data
US7925610B2 (en) Determining a meaning of a knowledge item using document-based information
US20120078891A1 (en) Providing answers to questions using multiple models to score candidate answers
US6611825B1 (en) Method and system for text mining using multidimensional subspaces
US20100332511A1 (en) System and Methods for Units-Based Numeric Information Retrieval
US20050108200A1 (en) Category based, extensible and interactive system for document retrieval
US8204751B1 (en) Relevance recognition for a human machine dialog system contextual question answering based on a normalization of the length of the user input
US6189002B1 (en) Process and system for retrieval of documents using context-relevant semantic profiles
US20100205198A1 (en) Search query disambiguation
US7421418B2 (en) Method and apparatus for fundamental operations on token sequences: computing similarity, extracting term values, and searching efficiently

Legal Events

Date Code Title Description
AS Assignment

Owner name: NEC LABORATORIES AMERICA, INC., NEW JERSEY

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:BAI, BING;WESTON, JASON;COLLOBERT, RONAN;REEL/FRAME:022189/0407

Effective date: 20090202

AS Assignment

Owner name: NEC CORPORATION, JAPAN

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:NEC LABORATORIES AMERICA, INC.;REEL/FRAME:031998/0667

Effective date: 20140113

FPAY Fee payment

Year of fee payment: 4

AS Assignment

Owner name: NEC CORPORATION, JAPAN

Free format text: CORRECTIVE ASSIGNMENT TO CORRECT THE REMOVE 8538896 AND ADD 8583896 PREVIOUSLY RECORDED ON REEL 031998 FRAME 0667. ASSIGNOR(S) HEREBY CONFIRMS THE ASSIGNMENT;ASSIGNOR:NEC LABORATORIES AMERICA, INC.;REEL/FRAME:042754/0703

Effective date: 20140113